Fix misuse of UCSConvertCase in XConvertCase

There are two issues with the use of UCSConvertCase in XConvertCase:

• Some Latin-1 keysyms do not map within Latin-1, e.g. ssharp. Only Latin-1 keysyms have the same value as the Unicode code point of their corresponding character. So UCSConvertCase does not work for some Latin-1 keysyms as it returns Unicode code points outside Latin-1 that do not match their corresponding keysyms values.
• Some Unicode keysyms should map to Latin-1 keysyms (<0x100). But the Unicode keysym mask 0x01000000 is applied blindly to the result of UCSConvertCase, resulting in invalid Unicode keysyms (0x010000nn) while they should be Latin-1 keysyms.

Example with ß/ẞ:

KeySym lower, upper;

XConvertCase(XK_ssharp, &lower, &upper);
// Expected: lower == XK_ssharp, upper == U1E9E
// Got:      lower == XK_ssharp, upper == 0x1E9E

XConvertCase(U1E9E, &lower, &upper);
// Expected: lower == XK_ssharp, upper == U1E9E
// Got:      lower == 0x10000df, upper == U1E9E

See #110 (closed) and !30 (merged).